Luyện nói tiếng Anh bằng Shadowing qua video: You NEED to try these 12 open-source AI projects RIGHT NOW

Over the weekend, I found some incredible open source AI projects that I wanted to share with you.

Let me show you them right now.

All right, the first one is called Open Montage and it turns your AI agent into a full video production team.

This one has almost 15,000 stars.

It describes itself as turning your AI coding assistant into a full video production studio.

Describe what you want in plain language.

Your agent handles research, scripting, asset generation, editing, and final composition.

and here's an example.

So this is a cinematic sci-fi trailer produced by Open Montage, created the concept, the script, the scene plan, used VO to generate the motion clips,

the soundtrack was generated and re-motion handled the composition.

You can also start from a video that you already love.

You simply give it the clip that you want

and it'll create a video very similar to that and of course you can change it any way you see fit.

And it has a ton of features including 12 production pipelines, explainer videos, talking heads, screen demos,

cinematic trailers, animation podcasts, localization, documentary montages, and more.

400 different agent skills and so much more.

I'm going to drop all the links for these down below.

And by the way, if you like these videos where I curate the best AI open source projects, drop a like, subscribe to the channel.

It very much does help.

Thank you in advance.

Next, we have a new agent harness coming out of ByteDance with possibly the weirdest name out of all of these projects.

It is called Deerflow, and it has almost 74,000 stars.

Deerflow stands for Deep Exploration and Efficient Research Flow.

It is an open source super agent harness that orchestrates subagents, memory, and sandboxes to do almost anything powered by extensible skills.

Deerflow is made for long horizon tasks.

That means you give the agent something and it goes off for hours, maybe even days at a time.

That is what this is especially good at.

It uses sub agents to break down complex tasks.

It has sandboxes.

Of course, it has memory and tools and skill use, everything you need, everything that you're familiar with, with all of the best agent harnesses out there.

But this one really leans into long horizon work.

So they talk about users using Deerflow for building data pipelines, generating slide decks, spinning up dashboards, automating content workflows.

So if you've tried OpenClaw, if you've tried Hermes, this might be another really cool option to experiment with.

All right, and next, and what likely will have to change its name, we have Anthropic Cybersecurity Skills coming in right below 20,000 stars.

This gives your agent everything it needs to be a cybersecurity expert.

Then you can point it at your code base and say, improve my cybersecurity defenses.

It works with Cloud Code, GitHub Copilot, Codex CLI, Cursor, Gemini CLI, basically anything.

Any agent that supports skills, you can plug this in.

So it comes with six different cyber frameworks.

I have not actually heard of these.

I'm not deep in the world of cyber, but we have the MITRE ATT&CK, NIST, and more.

All of these different frameworks, a bunch of different tactics within them, a bunch of different techniques within them,

just basically giving your agent a leg up and trying to protect your own apps that you're building.

Now, what's interesting about it is they actually took real cyber frameworks.

So, for example, this MITRE fight fraud framework co-developed by JPMorgan Chase, Citigroup, Lloyd's Banking Group, Standards Charter,

CrowdStrike, Verizon Business, and so on.

It basically puts together a framework for cybersecurity defense.

And of course, because it's only a skill, you simply copy paste the URL, put it in your agent and say, install it.

And it'll just work.

Next, we have an open source project from the company HeyGen coming in at just above 30,000 stars.

Hyper frames.

Hyper frames is an open source framework for turning HTML, CSS, media, and seekable animations into deterministic MP4 videos.

So basically it's going to write really cool animations for you in HTML, in CSS

and then it's going to convert that to video format, MP4.

This is really good for product demos, for slides, for motion graphics, and so much more.

It supports rendering in Chrome with FFmpeg and works with many animation libraries, including one of my favorite, 3JS.

So here's an example of what it looks like.

It has no sound, so I'm going to do my own voiceover.

Hey, I'm HyperframesGuy, and I'm going to tell you about Hyperframes.

Here are some cool Hyperframes logos and cool transitions.

Here's a prompt I'm writing.

You can ask your agent anything.

And there we go.

And I know all of these open source research projects are awesome, but they need to be powered by AI and all of those subscriptions become very expensive.

That's why I'm excited to tell you about the sponsor of today's video, Merlin AI.

Merlin AI is an all-in-one AI tool and they are giving my audience, you guys, a massive discount.

So stick around to the end of this read and I'll give you that code.

So you probably keep multiple AI tools on deck at all times because you'll have one tool that's good at one thing, another tool that's better at another,

and it can get expensive.

Merlin AI puts all the top LLM models from ChatGPT to Claude to Gemini all in one place.

So check this out.

I click the Merlin AI extension.

I chat with the webpage to summarize what I'm reading and I pull the important parts.

And if I need to go deeper, I just flip the switch on deep research.

And it also has the top video and image generation models ready.

So let's say you're going to pay for everything separately.

ChaiTQBT, 20 bucks.

Claude, 20 bucks.

Gemini, 20 bucks.

And it adds up fast.

Merlin AI is cheaper because they buy all of it via the API in bulk, and then they pass along those discounts to you.

So essentially, with Merlin's discount, it is only five bucks.

Okay, so here's the 75% off Merlin AI code.

It is MAT5.

Apply the code, and then the total drops to $60 per year.

Basically, five bucks a month.

So go try out Merlin AI.

I'll drop all of this in the description below.

Now, back to the video.

All right, so this next one is super interesting and a little bit under the radar, coming in just above 12,000 stars.

This is called Codebase Memory MCP by Deus Data.

This is described as the fastest and most efficient code intelligence engine for AI coding agents.

It full indexes an average repository in milliseconds, the full Linux kernel, 28 million lines of code in three minutes,

and then you can answer structural queries in under one millisecond.

This is meant to give your agent a better understanding of your code base, massive code bases, and to be hyper efficient.

It supports 158 different coding languages.

It uses 120 times fewer tokens.

It works in 11 different agentic harnesses, including all your favorites, Cloud Code, Cursor, Codex, etc.

It has a built-in 3D visualization so you can actually explore your code base

100

and see how it looks in this 3D graph space.

101

And it is one line to install.

102

Very easy to use.

103

Go check it out.

104

All right, next coming in at 143,000 stars, one of the most popular repositories in all of GitHub,

105

we have Matt Pocock's skills.

106

These are skills to give your agent to allow them to develop software just like Matt Pocock does.

107

And if you don't know who Matt Pocock is, this is Matt Pocock, author of Total TypeScript and AI Hero.

108

He was previously at Vercel and he is a developer educator.

109

He basically teaches you how to write great code.

110

So he took everything he learned throughout his entire career, built it into skills so you can give it to your agent and you can code just like him.

111

My agent skills that I use every day to do real engineering, not vibe coding.

112

Here's some of the features that come with the skills.

113

We have ask Matt, ask which skill or flow fits your situation or router over the user invoke skills in this repo.

114

Grill with docs, grilling session that also builds your project's domain model, sharpening terminology and updating context.md and ADRs inline.

115

This is for real engineers and obviously it's being used quite a bit because it has so many GitHub stars.

116

Go check it out again, links down below to all of these projects.

117

Next, we have the very popular GStack by Gary Tan,

118

president of Y Combinator, coming in at 114,000 stars.

119

GStack turns your engineering agent into a full engineering team.

120

This is how Gary thinks about building things.

121

And he basically codified all of the lessons he's learned throughout his career into G-Stack so you can easily use it, give it to your agent,

122

and allow them, your agents, to build things the same way that Gary Tan does.

123

He helps you not only with the building, but also the thinking through of different problems that you might want to solve.

124

This is especially valuable if you're thinking about building a startup.

125

You can do things like run slash office hours, which is something they do at Y Combinator.

126

You sit with a YC partner and they ask you a bunch of questions

127

and then give you feedback about the problem space, your solution, your team, everything.

128

And again, he just codified all of it and put it into G-Stack.

129

Installation is super simple.

130

It is just a set of skills.

131

And so you just copy paste the URL and your agent will know how to ingest it.

132

So he describes G-Stack as a process, not a collection of tools.

133

You are supposed to run these skills in order.

134

Think, plan, build, review, test, ship, reflect, and you do so with all of these different schools.

135

So we have slash office hours, slash plan CEO review, plan end review, plan design review, all the way through QA,

136

pair agent, CSO, ship, land and deploy, canary, benchmark, document release, everything, it handles it all end to end for you.

137

It is truly awesome.

138

Next, we have an open source OpenWeights vision language model from Baidu.

139

This one is brand new, coming in just under 3,000 stars.

140

Released just a couple days ago, you download this OpenWeights model, and all of a sudden, you have lightning fast OCR at your fingertips.

141

And if you're not familiar with what OCR is, it basically just means reading and analyzing documents.

142

So here are a few examples of what you can do.

143

On the left side, what you see is a research paper.

144

And what this project and model are doing is actually highlighting a PDF.

145

And if this doesn't sound impressive, let me tell you, this is actually a very hard problem.

146

You not only have to fully understand what's on the page, but you have to understand where on the page it is and how to highlight it appropriately.

147

And it is super fast, as you're seeing here.

148

Go to Hugging Face to download the actual model.

149

Click over here in Files.

150

And the model is only about six and a half gigabytes in size.

151

So very small.

152

Awesome, awesome job.

153

All right, next, an open source project out of NVIDIA.

154

And I should have probably put this one at the top of the video, but this one scans skills before you install them to make sure they are secure and safe.

155

This one is coming in under 10,000 stars.

156

It describes itself as a security scanner for AI agent skills, detect vulnerabilities, malicious patterns, and security risks.

157

And it's from NVIDIA, so you know, you can trust it.

158

So it comes with multi-format input.

159

You can scan, get repos, URLs, zip files, directories, or single files it looks for 65 vulnerability patterns across 16 categories,

160

including prompt injection, data exfiltration, privilege escalation, supply chain, excessive agency, output handling, and more.

161

Anytime you're about to install a skill, you should first use Skill Spectre to inspect it.

162

All right, next, this one is really cool.

163

It's called Palmier Pro.

164

It's at 8,000 stars, and it is a full AI native video editor that you can download for macOS.

165

Right now, it's only macOS.

166

Hopefully, they release it for Windows and Linux later, but right now, you download it, and the best part is, again, it's open source.

167

So, it has a full MCP server, built-in generative AI.

168

It integrates with your existing agents, including Cloud, Codex, Cursor, via MCP,

169

so you can control the downloaded video editor with your agent.

170

So you can tell it exactly how you want it to edit your videos and it will control this video editor.

171

It is beautiful.

172

And the best part is it's absolutely free.

173

And next, truly one of the most popular repos on all of GitHub, we have the Hermes agent coming in above.

174

Congratulations to them.

175

200,000 stars.

176

And boy, it seems like everybody is using it now.

177

It is a great alternative to OpenClaw.

178

Nothing wrong with OpenClaw.

179

This is just a very cool extra option that you can try out.

180

It has all the features and functionality that you're familiar with if you've ever used OpenClaw.

181

And it really leans into the self-healing, self-improving functionality.

182

So as you're using Hermes Agent, if one of the skills fails, it will automatically fix itself and improve for the next run.

183

It has a million features, so definitely go try this out.

184

I would need a full video to review this in full.

185

All right, next, we have VoiceBox coming in at 33,000 stars.

186

This promises to be both 11 labs and WhisperFlow.

187

So think about voice output, AI generated voice output, and also voice transcription.

188

So voice input, put them together and you have both sides of voice and it's all open source and it's all free.

189

You can even plug in local models and run it completely on your computer really well.

190

Actually, they sound quite good.

191

So clone any voice, generate speech, dictate into any app, talk to agents and voices you own the full voice IO stack running locally on your machine.

192

Here's an example of what the interface looks like.

193

Here you can type out what you want said.

194

It's using Quine 1.7b and you can see it's generating right there.

195

And it's a beautiful interface, very easy to use, very easy to edit your voice outputs.

196

And it has a bunch of features.

197

It is a very complete product.

198

So it has near perfect voice cloning.

199

Thank you to whatever model you're using.

200

It has a stories editor so you can actually edit the audio like you would in any timeline editor.

201

You have audio effects pipeline, local or remote.

202

We have audio transcription and unlimited generation all coming to you within this product.

203

And I actually made a video going over four incredible open source projects in depth.

204

I actually show you how to install them, show you how to use them, check it out right here.

Phương Pháp Shadowing Là Gì?

Shadowing là kỹ thuật học ngôn ngữ có cơ sở khoa học, ban đầu được phát triển cho chương trình đào tạo phiên dịch viên chuyên nghiệp và được phổ biến rộng rãi bởi nhà đa ngôn ngữ học Dr. Alexander Arguelles. Nguyên lý cốt lõi đơn giản nhưng cực kỳ hiệu quả: bạn nghe tiếng Anh của người bản xứ và lặp lại to ngay lập tức — như một "cái bóng" (shadow) đuổi theo người nói với độ trễ chỉ 1–2 giây. Khác với luyện ngữ pháp hay học từ vựng bị động, Shadowing buộc não bộ và cơ miệng phải đồng thời xử lý và tái tạo ngôn ngữ thực tế. Các nghiên cứu khoa học xác nhận phương pháp này cải thiện đáng kể phát âm, ngữ điệu, nhịp điệu, nối âm, kỹ năng nghe và độ lưu loát khi nói — đặc biệt hiệu quả cho người luyện IELTS Speaking và muốn giao tiếp tiếng Anh tự nhiên như người bản ngữ.

Luyện nói tiếng Anh bằng Shadowing qua video: You NEED to try these 12 open-source AI projects RIGHT NOW

Ngữ Cảnh & Nền Tảng

5 Cụm Từ Chính Trong Giao Tiếp Hằng Ngày

Hướng Dẫn Shadowing Từng Bước

Phương Pháp Shadowing Là Gì?